Data Extension Device for Machine Learning Time Series Regularity

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Sufficient data extension cannot be achieved when the starting and ending dates/times of learning execution data are uniquely defined, limiting the ability to improve generalization performance of machine learning models.

Innovation Solution

A data extension method that combines minimum constitution unit data from entire learning data, assigning labels to maintain regularity, allowing for the generation of learning execution data that maintains the regularity of the original time series, even when starting and ending dates/times are fixed.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If the starting date/time and ending date/time of learning execution data are uniquely defined, then the data structure is simple and clear, but the number of times of shifting the starting date/time is reduced, and sufficient data extension cannot be achieved

Engineering Contradiction:

Improvedata structure definition precisionVSAvoiddata extension capability

Core Design Contradiction:

Manufacturing precisionVSAdaptability or versatility

Solution Approach 1:

The patent segments the learning execution data into multiple minimum constitution unit data sets, each representing a fixed time interval (e.g., 90 minutes). By dividing the overall time series into discrete segments, the system can systematically shift and combine these segments to generate multiple learning execution data sets while maintaining the fixed interval structure. This segmentation enables data extension without compromising the clarity of the data structure definition.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent combines multiple minimum constitution unit data sets to form complete learning execution data sets. By merging segmented data units with shifted time intervals, the system generates sufficient training data while preserving the fixed starting and ending date/time definitions. This combining approach resolves the contradiction by enabling data extension through structured composition rather than arbitrary time shifting.

Inventive Principle:

Principle #5Merging (Combining)

2Quantity of substance

If the starting date/time is shifted frequently to achieve data extension, then the number of learning data increases, but the time interval between data points becomes irregular, affecting the regularity of the time series

Engineering Contradiction:

Improvenumber of learning dataVSAvoidtime series regularity

Core Design Contradiction:

Quantity of substanceVSStability of the object's composition

Solution Approach 1:

The patent applies local quality by maintaining different characteristics for different parts of the data processing: the minimum constitution unit data sets maintain fixed time intervals and regular structure, while the combination of these units through shifting creates diverse learning execution data sets. This local differentiation allows frequent shifting for data extension while preserving time series regularity within each constituent unit.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent implements periodic action by systematically shifting the starting date/time of minimum constitution unit data sets by fixed intervals (e.g., shifting by the length of one minimum constitution unit). This periodic shifting pattern generates multiple learning execution data sets while maintaining regular time series structure, as each shift follows a consistent periodic rule rather than arbitrary timing.

Inventive Principle:

Principle #19Periodic action

3Device complexity

If the time interval of learning execution data is fixed, then the data structure is simple, but the ability to adapt to different learning requirements is limited

Engineering Contradiction:

Improvedata structure complexityVSAvoidlearning requirement adaptability

Core Design Contradiction:

Device complexityVSAdaptability or versatility

Solution Approach 1:

The patent introduces dynamics by allowing the learning execution data to be flexibly constructed from multiple minimum constitution unit data sets through systematic shifting. While each individual unit maintains a fixed time interval for structural simplicity, the overall system becomes adaptable to different learning requirements by adjusting the number of shifts, the combination of units, and the specific time intervals used in generating training data sets.

Inventive Principle:

Principle #15Dynamics

Data Source

PatentUS20220391715A1Data extension device, data extension method, and data extension program

Publication Date: 2022.12.08 NIPPON TELEGRAPH & TELEPHONE CORP

AI summary

Sufficient data extension can be achieved even if the starting date/time and ending date/time of the learning execution data have been uniquely defined. Based on entire learning data that is a set of time-series data, wherein the entire learning data is a set of minimum constitution unit data, each of which is time-series data having a first time interval that is a time interval required for learning, wherein the time-series data having the first time interval is each assigned a first label that indicates a feature in a time series of the first interval; a generation unit (103) generates learning execution data that is a set of time-series data to be used in learning, by combining the minimum constitution unit data included in the entire learning data such that regularity of the first label in a time series of the entire learning data is maintained.